Figure 2
From: Event detection in consumer videos using GMM supervectors and SVMs
System overview. In the feature extraction phase, eight types of features are extracted for each clip. CC-DSTIP is our new feature. In the modelling phase, GMM supervectors corresponding to eight features are constructed. In the detection phase, the GMM supervector for each feature is used as the input to an SVM, and finally, SVM scores are fused by weighted average. If the fused detection score is higher than the threshold, the clip is detected as an event and otherwise not.